[TF] Removed 'InitTensorFlowRuntime'

stdlib/public/CTensorFlow/ctensorflow_init.cpp

@@ -17,33 +17,6 @@ void handle_sigint(int signal) {
   exit(1);
 }
 
-void InitTensorFlowRuntime(unsigned char enable_debug_logging,
-                           int verbose_level) {
-  // Install a signal handler to ensure we exit when interrupted.
-  signal(SIGINT, handle_sigint);
-
-  // Synthesize argc and argv
-  char arg0[] = "dummyProgramName";
-  std::vector<char*> my_argv;
-  my_argv.push_back(&arg0[0]);
-  // This allows us to dump TF logging to the output of a swift binary.
-  // We can only dump to stderr, since there is no flag alsologtostdout.
-  char arg1[] = "--alsologtostderr";
-  if (enable_debug_logging > 0) {
-    my_argv.push_back(&arg1[0]);
-  }
-  char arg2[] = "--v=?";
-  if (verbose_level > 0) {
-    assert(verbose_level <= 4);
-    arg2[4] = verbose_level + '0';
-    my_argv.push_back(&arg2[0]);
-  }
-  int my_argc = my_argv.size();
-  char** tmpArgv = my_argv.data();
-  // Initialize GPU devices.
-  TF_InitMain(/*usage=*/nullptr, &my_argc, &tmpArgv);
-}
-
 static bool setValue(TF_DataType tfDtype, int64_t val, void *ptr) {
   switch (tfDtype) {
   case TF_INT8:

stdlib/public/CTensorFlow/ctensorflow_init.h

@@ -7,15 +7,6 @@
 extern "C" {
 #endif
 
-// Call this API exactly once before any TensorFlow backend/runtime calls.
-//
-// It sets up device context for any GPU based computation.  When
-// `enable_debug_logging` is true, it also dumps TF logging for debugging
-// purposes. In that case, when `verbose_level` is positive (must be <= 4), it
-// also dumps verbose logs at that level.
-extern void InitTensorFlowRuntime(unsigned char enable_debug_logging,
-                                  int verbose_level);
-
 //===----------------------------------------------------------------------===//
 // - MARK: Runtime functions to be called via IRGen.
 //===----------------------------------------------------------------------===//

stdlib/public/TensorFlow/CompilerRuntime.swift

@@ -580,8 +580,44 @@ public final class _ExecutionContext {
     // Initialize the TF runtime exactly once. Only affects local execution
     // (when _RuntimeConfig.tensorFlowServer is set to "").
     if !_RuntimeConfig.tensorFlowRuntimeInitialized {
-      InitTensorFlowRuntime(_RuntimeConfig.printsDebugLog ? 1 : 0,
-                            _RuntimeConfig.tensorflowVerboseLogLevel)
+      var args = ["dummyProgramName"]
+      if _RuntimeConfig.printsDebugLog {
+        args.append("--alsologtostderr")
+      }
+      if _RuntimeConfig.tensorflowVerboseLogLevel > 0 {
+        args.append("--v=\(_RuntimeConfig.tensorflowVerboseLogLevel)")
+      }
+      // Collect all the strings' utf8 bytes into a single array so that we can
+      // address all the strings with a single `flattenedStringBytes.withUnsafeBufferPointer`.
+      var flattenedStringBytes: [Int8] = []
+      var lengths: [Int] = []
+      for arg in args {
+        let bytes = arg.utf8CString
+        flattenedStringBytes.append(contentsOf: bytes)
+        lengths.append(bytes.count)
+      }
+
+      // Calculate the addresses of all the strings within our single buffer, and then call
+      // TF_InitMain.
+      flattenedStringBytes.withUnsafeMutableBufferPointer { flattenedStringBytesBuffer in
+        var stringAddrs: [UnsafeMutablePointer<Int8>?] = []
+        var currentStringAddr = flattenedStringBytesBuffer.baseAddress.map(UnsafeMutablePointer.init)
+        for length in lengths {
+          stringAddrs.append(currentStringAddr)
+          currentStringAddr = currentStringAddr?.advanced(by: length)
+        }
+
+        stringAddrs.withUnsafeMutableBufferPointer { stringAddrsBuffer in
+          var cArgs = [stringAddrsBuffer.baseAddress.map(UnsafeMutablePointer.init)]
+          var cArgsCount = [Int32(args.count)]
+
+          cArgs.withUnsafeMutableBufferPointer { cArgsBuffer in
+            cArgsCount.withUnsafeMutableBufferPointer { cArgsCountBuffer in
+              TF_InitMain(nil, cArgsCountBuffer.baseAddress, cArgsBuffer.baseAddress)
+            }
+          }
+        }
+      }
       _RuntimeConfig.tensorFlowRuntimeInitialized = true
     }